As a point, at least in my implementation, I actually use a byte stream for the 1x1 that is cached and sent to the browser. We do this so that there is no disk

As a point, at least in my implementation, I actually use a byte stream for the 1x1 that is cached and sent to the browser. We do this so that there is no disk read ever for it. We also don't use php/gd, our code is mostly node.js and a little golang with nginx as our front end. It may seem a little overkill, but I found doing it this way made a pretty significant impact when you have a reasonably decent volume of tracking going on. A good http server of course would cache the image of course, but we just found this to be easier and more consistent for our implementation.